Dwi Nugroho from TU Berlin will visit the DASYA research group and give a talk at ITU on December 11, 2026, between 11:00–12:00, in room 3A08.

Efficient Stream Processing on Heterogeneous Hardware

Abstract: Stream processing systems must handle growing workloads while delivering timely results at manageable costs. Hardware accelerators offer additional processing capacity, but their benefits depend on workload characteristics and environment dynamics. In this talk, I present three contributions toward effective stream processing on heterogeneous hardware. First, I examine when stream join algorithms benefit from GPU acceleration. Second, I introduce a model for exploring design choices under throughput, latency, and resource constraints. Third, I present an adaptive and contextual optimizer that adjusts execution as workloads and concurrent queries change. Together, these contributions provide practical guidance for designing and operating stream processing systems that use hardware accelerators efficiently.

Speaker: Dwi Nugroho